TRIAGE: Dialectical Reasoning for Explainable Risk Prediction on Irregularly Sampled Medical Time Series with LLMs

2026-06-08 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, AI in Healthcare · Depth: Expert, quick

Summary

TRIAGE is a novel framework designed for explainable risk prediction using Large Language Models (LLMs) on irregularly sampled medical time series (ISMTS). It addresses a critical limitation where existing LLM approaches for clinical early warning systems often collapse graded clinical risk into overconfident binary predictions, undermining both calibration and cross-patient comparability. TRIAGE mitigates this risk polarization by training an LLM to generate dialectical reasoning, eliciting outcome-specific rationales for competing clinical outcomes. This enables the LLM to yield continuous risk scores grounded in explicit clinical reasoning. Benchmarked on three ISMTS datasets, TRIAGE demonstrated an average AUPRC improvement of 3.3% and reduced calibration error by 81% compared to competitive baselines. Furthermore, an LLM-as-a-judge assessment confirmed its rationales surpassed post-hoc explanations from baselines by 20% in clinical reasoning quality.

Key takeaway

For Machine Learning Engineers developing clinical early warning systems, TRIAGE offers a robust approach to overcome LLM limitations in risk prediction. If you are struggling with overconfident binary predictions or poor calibration in medical time series analysis, consider implementing TRIAGE's dialectical reasoning framework. This method provides continuous, calibrated risk scores and high-quality, verifiable clinical rationales, significantly improving model interpretability and trustworthiness for patient triage.

Key insights

TRIAGE uses dialectical LLM reasoning to provide calibrated, continuous risk scores and interpretable rationales for medical time series.

Principles

• LLMs can generate dialectical reasoning.
• Outcome-specific rationales mitigate risk polarization.
• Continuous risk scores require explicit clinical reasoning.

Method

TRIAGE trains an LLM to generate dialectical reasoning by eliciting outcome-specific rationales for competing clinical outcomes, enabling continuous risk scores grounded in explicit clinical reasoning.

In practice

• Apply dialectical reasoning to LLM risk prediction.
• Use outcome-specific rationales for better calibration.
• Integrate TRIAGE for explainable medical early warnings.

Topics

• TRIAGE Framework
• Large Language Models
• Medical Time Series
• Risk Prediction
• Explainable AI
• Clinical Early Warning Systems

Code references

• HyeongWon-Jang/TRIAGE

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Please don’t trust your chatbot for medical advice — LLMs are unreliable for medical advice, frequently generating misinformation with overconfidence and lacking clinical reasoning.
• Clinically Interpretable Sepsis Early Warning via LLM-Guided Simulation of Temporal Physiological Dynamics — LLM-guided temporal simulation improves sepsis early warning by providing interpretable physiological trajectory predictions.
• Statistics, Not Scale: Modular Medical Dialogue with Bayesian Belief Engine — Strictly separating language processing from probabilistic reasoning in medical AI yields superior, auditable, and private diagnostic systems.
• ClinEnv: An Interactive Multi-Stage Long Horizon EHR Environment for Agents — ClinEnv reveals LLMs struggle with sequential, uncertain clinical decisions and information gathering in a multi-stage environment.
• PSEBench: A Controllable and Verifiable Benchmark for Evaluating LLMs in Patient Safety Event Triage — PSEBench offers a verifiable benchmark for LLMs in patient safety triage via policy-grounded construction and agentic evaluation.
• An Infectious Disease Spread Simulation Based on Large Language Model Decision Making — LLMs can simulate nuanced human health behaviors in spatially-grounded agent-based models, revealing demographic disparities.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.